Claude Code 2.1.257 closes several ways a permission rule could be walked around
Anthropic shipped two Claude Code releases on September 1, 2026. Version 2.1.257 adds a containment-escape rule to auto mode and a prompt before the first file read outside the working directories, and it fixes five places where a differently written command slipped past a deny rule. A project settings file can no longer put a session into bypass mode. Version 2.1.258 restores launching on macOS 12, broken since 2.1.255.
The release notes for Claude Code 2.1.257, published on September 1, 2026, read as a pass over the permission system rather than a feature list. Auto mode gained a rule aimed at containment escapes, five deny rules turned out not to cover the commands they were written for, and a file checked into a repository can no longer switch a whole session into bypass mode.
What auto mode stops approving on its own
The new Containment Escape rule covers three moves: fetching cloud metadata credentials, evading egress controls, and reaching across tenants. None of them are auto-approved any more unless the environment marks them as expected. These are the actions an agent takes when it is trying to get out of the box it was put in, whether it got there by itself or because a page it read told it to.
The second change is quieter and matters more day to day. Auto mode now shows a one-time prompt before the first file read outside the working directories, and permissions.blockReadsOutsideWorkingDirectories can refuse such reads outright. Until now a session could read anything the account could read, and nothing in the interface said so.
Deny rules that were not denying
Five fixes in this release have the same shape. A permission rule existed, and one particular way of writing the command went past it.
One setting changed meaning as well. defaultMode: "bypassPermissions" in a project's .claude/settings.json or .claude/settings.local.json is now ignored, the same way "auto" already was. A file committed to a repository can no longer decide that everyone who opens it works without permission prompts; that belongs in user settings, in managed settings, or on the command line.
- A
permissions.askrule was skipped in auto mode when the matching command ran inside a compound command or a subshell, so it ran with no prompt at all. - Bash
Read()andEdit()deny rules did not apply to< fileredirects or to reader commands such astacandegrep; a deny rule on any argument or redirect target now refuses the command. - A sandbox network host written with a trailing dot,
example.com., was not blocked by the matchingdeniedDomainsentry. - Dismissing the Remote Control consent prompt counted as consent, so the next request connected without asking again.
- A plugin could read files outside its own directory through a declared command, agent, skill or hook path that was a symlink; such paths are now refused with an error.
Fable 5.1 as the default, and a launch that had broken
The same release adds Claude Fable 5.1 as the default model in that line, with a one-million-token context window. It is priced at 10 dollars per million input tokens and 50 dollars per million output tokens, with cache reads at 0.25 dollars per million tokens. In Claude apps gateway sessions, fable and best keep resolving to Fable 5 for now, because gateways not yet configured for the new model reject it.
Version 2.1.258 followed a few hours later with two fixes that restore basic operation. Claude Code was failing to launch on macOS 12, a regression introduced in 2.1.255, and remote and scheduled sessions were failing after a re-sent permission approval could not be applied.
Sources
Related
OpenClaw 2026.10.2 stops long tool output from slipping credentials past redaction
A bug in OpenClaw's secret masking let a credential through whenever it straddled the boundary between two slices of a long text. The fix ships in 2026.10.2 along with repaired updates from 2026.10.1 and a new compaction default for OpenAI and Anthropic API keys.
OpenClawverified

OpenClaw 2026.10.1 lets agents search installed skills and changes how sessions message each other
OpenClaw released version 2026.10.1 on October 10, 2026, its first stable release since 2026.9.7. Agents can now search the skills installed on a machine and read only the one a task needs, instead of carrying the whole library in the prompt. Opt-in MCP Apps run beside chat, and full backups can go to an external drive or Cloudflare R2. Three changes can break existing setups: session-to-session messaging, the old required argument, and Workboard data stored before July 2026.
OpenClawverified
Claude Code 2.1.294 fixes prompt hooks that let through the actions they were written to block
Anthropic released Claude Code 2.1.294 on October 8, 2026. It fixes prompt and agent hooks phrased as instructions, which could allow exactly the actions they were meant to stop. Version 2.1.295 followed the same day with an onFailure setting that makes a crashed or timed-out command or HTTP hook block the action. Releases 2.1.292 and 2.1.293 from earlier in the week closed several permission and sandbox gaps.
Claude Code Releases (GitHub)verified